2022 Fiscal Year Final Research Report
Hidden Markov Models for onset of type 2 diabetes
Project/Area Number |
19K17970
|
Research Category |
Grant-in-Aid for Early-Career Scientists
|
Allocation Type | Multi-year Fund |
Review Section |
Basic Section 54040:Metabolism and endocrinology-related
|
Research Institution | Aichi Medical University |
Principal Investigator |
|
Project Period (FY) |
2019-04-01 – 2023-03-31
|
Keywords | 糖尿病予防 / 健康診断 / 機械学習 / 前糖尿病 |
Outline of Final Research Achievements |
In this study, a hidden Markov model was applied to workers' health examination data to construct a model of diabetes development. The model determined the unwell (healthy) state, the high-risk state and the diabetes state, and assumed that there is an annual back and forth (transition) from state to state. The results of the model construction suggested that the estimated distribution of the 'high-risk state' (104.6 ± 7.1 mg/dl) suggests that the WHO cut-off point of 110 mg/dL for the high-risk of type 2 diabetes may be too high. Furthermore, the estimated probability of transition from 'high risk state' to 'normal state' was very low at 0.01%, suggesting that those who have reached 'high risk state' are unlikely to return to 'normal state'.
|
Free Research Field |
産業保健
|
Academic Significance and Societal Importance of the Research Achievements |
健康診断縦断データに機械学習モデルを当てはめた結果、これまでの「ハイリスク」基準が高すぎること、「ハイリスク状態」に至った者は「正常状態」に戻ることはほとんどない可能性が示唆された。したがって2型糖尿病予防のためには、特定保健指導の文脈とは別に「ハイリスク状態」として特定された個人に対する毎年のモニタリングと継続的な介入が必要であること、「ハイリスク状態」に至らないための「正常状態」の者たちへのポピュレーションアプローチが重要になることが推察された。 このように、機械学習による2型糖尿病発症モデルは、発症の実態と予防戦略についての理解を向上させる可能性がある。
|